Will be made available upon project completion. Further details are in the Pre-Analysis Plan.
Intervention
Exogenous treatment variation in people's income and transfer levels. Further details are in the Pre-Analysis Plan.
Experimental design
The survey will be distributed to a representative sample of US adults. Each survey respondent will take the same survey, but the transfer choice sets will be randomized.
Design details
Randomization method
Randomization done when the survey loads, by a computer.
Randomization unit
Randomization is at the respondent-question level.
Clusters
Clusters are respondents; standard errors are clustered at the respondent level. Wave 1: 2,421 respondents. Wave 2: 2,456 respondents. Wave 3: 1,232 respondents. Wave 4: 3,103 respondents. Wave 5: 2,000 respondents. Total: approximately 11,200 respondents.
Observations
Wave 1: 10 pairwise transfer choices per respondent (approximately 24,000 choice observations). Wave 2: 10 pairwise transfer choices per respondent (approximately 25,000 choice observations). Wave 3: 10 transfer choices per respondent plus stated average tax rates (approximately 12,000 choice observations). Wave 4: 10 pairwise transfer choices per respondent (approximately 31,000 choice observations). Wave 5: 15 single-family tax-price choices per Arm 1 respondent, one policy-schedule choice per Arm 2 respondent, and 10 pairwise transfer choices per Arm 3 respondent, plus stated average tax rates (approximately 18,000 choice observations). Standard errors are clustered at the respondent level.
Arms
Wave 1 (N = 2,421): no discrete arms (continuous randomization of choice sets). Wave 2 (N = 2,456): four information conditions cross-randomized with a norms treatment, ~307 per cell. Wave 3 (N = 1,232): two groups of ~616 (standard transfers plus stated tax rates; transfers taxed in favor of the U.S. Treasury). Wave 4 (N = 3,103): four framing conditions, ~776 each. Wave 5 (N = 2,000): three arms of ~500, 500, and 1,000; further between-subject randomizations within arms.
Primary outcomes
The primary outcome of the experiment is an estimated distribution of redistribution preferences. See PAP for further details.
Update January 14 2026:
In wave 4, we will run the following regression which is already part of the current (public) draft which analyzes the previous waves: Y_{iq} = β₁ [τ_{iq1} − τ_{iq2}]
+ β₂ (C_{iq1} + C_{iq2}) / 2
+ β₃ [C_{iq1} − C_{iq2}]
+ β₄ (T_{iq1} + T_{iq2}) / 2
+ β₅ [T_{iq1} − T_{iq2}]
+ ε_{iq}
This regression will be run 4 times, one for each experimental group. We will further interact regressors with efficiency beliefs and attitudes toward government that we elicit at the end of the survey.
We will also also plot stated governemnt inefficiency beliefs by political identity (democrat, republican, moderate).
Wave 5: (i) the probability of choosing transfer options that involve donations to the U.S. government, as a function of the randomized tax rate and the family's income (Arm 1); (ii) the probability of choosing each of three redistributive tax policies, and how it responds to labeling the status-quo policy (Arm 2); (iii) the choice-probability regression specified in the January 14, 2026 update, estimated by randomized factor cell (stakes, horizon, framing, group size) and with factor interactions (Arm 3), and the probability of transferring to zero-income families with and without the randomized unable-to-work description. Stated preferred average tax rates are anotheroutcome.
Secondary outcomes
- Participants' subjective beliefs about tax elasticities.
- Stated preferred average tax rates where not listed as a primary endpoint (waves 4 and 5).
- Beliefs about the government (e.g., what the government would do with donated funds, whether donating to the government resembles paying federal taxes) and the attitudes-toward-government battery (waves 4 and 5).
- Interactions of redistributive preferences with the vector of observables elicited in the survey. See attached survey for elicited observables.
- Narratives about why people make the transfer choices they have made in the survey. This is an open-end question where we will classify narratives into categories based on LLM and manual work by research assistants.
Cited by
No citation data available for this registration yet.